Mapping and revealing the nature of masonry compressive strength using computational intelligence

Abstract

The compressive strength of masonry walls constitutes a significant parameter that strongly influences the structural response of masonry buildings, under either static or dynamic actions. Significant variability is observed in the range of compressive strength values as highlighted by existing experimental investigations. Empirical relations providing the compressive strength also feature significant prediction divergence. This is attributed to large variations in the geometry and type of units, joint thicknesses, materials and building practices. Therefore, the need arises for the accurate prediction of the compressive strength of masonry walls, using data which is accumulated from past experiments. Artificial intelligence tools and machine learning techniques are considered in this study, to leverage the experience from those past experiments in predicting the compressive strength. A dataset of 611 specimens is developed, to the authors’ best knowledge comprises the largest dataset assembled for this purpose to date. Different Back Propagation Neural Networks models are trained and tested using the new dataset, leading to an optimal machine learning architecture. Results indicate that the optimal model can provide an improved prediction of the compressive strength as compared to literature proposals. Parameters which drastically affect the compressive strength are highlighted and expressions predicting the compressive strength are discussed.